One Simple System Gave All My AI Tools a Memory. Here's How.

Study Guide

Overview

Nate B. Jones builds on his original Open Brain concept — a personal Supabase database connected to AI tools via MCP — by showing how to add visual interfaces ("the human door") and practical extensions that make the system genuinely useful in daily life.

Key Takeaways

  • The keyhole problem: Chatting with AI through a text interface is powerful but limited. You need both an agent-readable interface (MCP) and a human-readable interface (visual app) pointing at the same data.
  • One source of truth, two doors: The database table is the shared surface. Your agent reads/writes via MCP; you read/write via a simple web app. No sync layer, no export, no middleware.
  • Build the visual layer with AI: Describe what you want to your AI (Claude, ChatGPT), get a web app generated, then deploy it for free on Vercel. Bookmark it on your phone and it behaves like a native app.
  • The AI flywheel: Every time models improve, your entire Open Brain system automatically gets smarter. You are just logging data and building patterns; the intelligence compounds over time.

Core Architecture

The Two-Door Pattern

  • Agent door: MCP server connected to your Supabase database — the agent queries, writes, reasons across tables
  • Human door: A visual web app (deployed on Vercel) that reads from and writes to the same tables
  • Both sides have direct access to the same rows — consistency is architectural, not managed by a sync layer

How to Build a Visual Extension

  1. Create the table and columns in your Supabase Open Brain instance
  2. Describe the visual you want to your AI (e.g., "mobile-friendly maintenance tracker with warranty dates highlighted")
  3. Iterate on the generated app with the AI
  4. Deploy to Vercel (free tier) to get a live URL
  5. Bookmark on your phone — it behaves like an app without the App Store

Use Cases

Household Knowledge Base

Capture scattered household info (paint colors, plumber contacts, Wi-Fi passwords, kids' shoe sizes) into a structured table during normal AI conversations. Build a searchable, categorized visual interface over it.

Professional Relationship Tracker

Log interactions with professional contacts. Ask your AI "anyone I've been neglecting?" and it scans for gaps. The visual shows at-a-glance which relationships need attention, with the ability to filter by topic or industry.

Job Search Dashboard

Track companies, roles, contacts, interviews, follow-ups, and resume versions across tables. Cross-category reasoning surfaces warm introductions from conference notes and relationship data. An autonomous agent can catch expiring follow-up windows you forgot about.

Three Principles for Choosing Problems

  1. Time bridging: Agent memory doesn't decay like human memory. Any problem where value comes from linking events across months or years is agent territory (maintenance schedules, recurring compliance tasks).
  2. Cross-category reasoning: The power is in connections between tables that humans don't naturally cross-reference (e.g., dishwasher maintenance status + meal planning).
  3. Agent surfaces, human decides, agent executes: The agent handles memory and pattern recognition; you handle judgment and decisions. This division keeps the system trustworthy.

Key Concepts

  • Open Brain: A personal Supabase database with an MCP server that any AI client can connect to
  • MCP (Model Context Protocol): The protocol that lets AI agents read/write to your database directly
  • Vercel: Free hosting service for deploying the visual web apps
  • Supabase: Open-source database platform where the Open Brain data lives
  • Agent-readable surface: Data structured so AI can query and reason about it, not just scrape a UI

Practical Advice

  • You don't need an autonomous agent (like Open Claw) to benefit — Claude or ChatGPT with MCP access already provides huge value
  • Start with Supabase's built-in table editor before building custom visuals
  • You can use Lovable as a shortcut for the visual layer, or build it yourself with AI-generated code for free
  • Once you deploy your first Vercel app, adding more is trivially easy
  • The pattern is the same regardless of domain: define the table structure, point your agent at it through MCP, point yourself at it through a visual interface
YouTube